Fuzzy Logic based Image Thresholding

Abstract

Thresholding is one of the most direct and computationally efficient approach to segment an image into its constituent regions and objects. In general, thresholding gives good results only when the intensity classes are separable and produces erroneous results in the presence of non-uniform illumination and noise in the image. In this thesis, a novel automated process is proposed for thresholding to reduce the error using non-linear filtering and fuzzy logic properties. The core idea is to apply adaptive nonlinear filters to suppress the noise, retaining the essential information and automate the process of thresholding. Being automated, there is no need of manual selection of the filter size. The starting point of threshold selection is to calculate centroids using clustering methods and make respective classes of intensity values. A membership function is then calculated for each class and the membership value is modified by considering the spatial information of the respective region pixels. Finally, an optimal threshold is obtained by defuzzification method. The performance of the proposed technique is evaluated, subjectively as well as objectively, by extensive experiments on different types of images. Further, the comparisons with existing techniques confirmed the effectiveness of the proposed technique.